Browse
Browse
2,352 records in 17 classes and 131 techniques.
Direct Prompt Injection 53
Instructions typed into the model that override what the application told it to do.
Indirect Prompt Injection 20
Instructions hidden in documents, web pages, emails or tool output that the model later reads.
Jailbreak and Guardrail Bypass 3
Getting a model past its safety rules or content filters.
System Prompt Leakage 2
Extracting hidden instructions or configuration from a model or agent.
Sensitive Data Disclosure 62
Secrets, personal data, chat history or training data exposed through an AI system.
Improper Output Handling 36
Model output trusted by downstream code: XSS, SQL injection, SSRF or code execution through what the model wrote.
Tool and Agent Misuse 68
An agent or chain runs code, commands or tools on behalf of an attacker.
Agent Context and Memory Poisoning 3
Planting content in an agent’s memory, history or context so later sessions are steered.
MCP and Tool Server Flaws 166
Vulnerabilities in Model Context Protocol servers, clients, plugins and agent tool integrations.
Unsafe Model Loading and Supply Chain 241
Malicious model files, pickle deserialisation, poisoned packages and artifacts.
Data and Model Poisoning 4
Tampering with training data or model weights, including backdoors.
RAG and Vector Store Weaknesses 36
Flaws in retrieval pipelines, embeddings and vector databases.
Model Evasion 17
Inputs crafted so a model misclassifies or fails to detect.
Model Theft and Extraction 4
Stealing, replicating or inverting a model through access to it.
Unbounded Consumption 117
Denial of service, resource exhaustion and cost attacks.
Misinformation and External Harm 1
Hallucinated or manipulated output that causes real-world harm.
AI Framework and Platform Flaws 1,519
Classic software bugs in the libraries and platforms that run AI: memory corruption, traversal, SSRF, auth.